feat(agents): add interactive streaming GAN REPL

[FluffyAIcode]

Summary

• add an interactive prompt> REPL that starts missing Gemma services and leaves them running
• warm each Generator/Critic context through allens prefill, then stream actual Primary hot-hit tokens to Terminal in real time
• print per-prompt KV hit rate, decode tok/s, generation ms/token, E2E tok/s and benchmark run id
• persist only redacted output hash/length and metrics to kakeya.ai; terminal alone shows model text
• support repeated prompts until /quit or EOF

Tests

• streaming suffix, cache gate and privacy report tests — 7 passed
• full platform-neutral Python suite — 850 passed, 1 skipped
• Python compile, shell syntax and diff checks pass

Live validation after merge

Open Terminal with bash scripts/run_agent_gan_repl.sh, enter a real prompt, verify token streaming for Generator and Critic, then inspect the resulting run in kakeya.ai Benchmarks.

Made with Cursor

[cursor]

Bugbot is not enabled for your account, so this pull request was not reviewed.

Enable Bugbot in the Cursor dashboard to get automatic reviews on future PRs.